- 6.1 About Ethernet
- 6.2 About Hubs, Switches, and Routers
- 6.3 About TCP/IP
- 6.4 About Packets
- 6.5 About Remote Procedure Calls (RPCs)
- 6.6 Slop
- 6.7 Observing Network Traffic
- 6.8 Sample RPC Message Definition
- 6.9 Sample Logging Design
- 6.10 Sample Client-Server System Using RPCs
- 6.11 Sample Server Program
- 6.12 Spinlocks
- 6.13 Sample Client Program
- 6.14 Measuring One Sample Client-Server RPC
- 6.15 Postprocessing RPC Logs
- 6.16 Observations
- 6.17 Summary
- Exercises
This chapter is from the book
This chapter is from the book
This chapter is from the book
6.14 Measuring One Sample Client-Server RPC
Recall our thought framework from Chapter 1. In thinking about performance issues, we will follow the programmer’s discipline of first estimating how long some work should take, then observing how long it actually does take, and then reasoning about any differences. Figure 6.15 shows this framework again.
)
Figure 6.15 Framework for examining the performance of complex software
Consider the offered load of sending a single write RPC from client program to server program, with a 5-byte key and 1,000,000-byte value. What do we expect the timing to look like on our sample server configuration with a 1Gb/sec network?
Remember that 1 Gb/sec is roughly 100 MB/sec when we convert from bits to bytes and include some overhead, or about 100 bytes per microsecond. So sending ~1,000,000 bytes of RPC request should take about 10,000 usec (10 msec) across the wire. On the server side, creating a string of 1,000,000 bytes and putting it into a C++ map should take only about 100 usec or so if the main memory can transfer 10GB/sec. A short reply of 100 bytes should take about 1 usec on the wire plus some small software overhead. Overall, we might expect something like Figure 6.16 when we sketch out a timeline.
)
Figure 6.16 Expected RPC timing sketch for Write() of 1MB of data
This sketch is somewhat distorted, but you get the idea—a long time to transmit 1MB in the request message, 100x less time to put it into the key/value store, and 100x less time again to transmit the response message. Let’s see what happens.